arXiv subjects

Yujing Wang

Publications and source records attributed to Yujing Wang.

64 records · Page 4Linked to original sources

arXiv2020

Automated Model Selection for Time-Series Anomaly Detection

Time-series anomaly detection is a popular topic in both academia and industrial fields. Many companies need to monitor thousands of temporal signals for their applications and services and require instant feedback and alerts for potential incidents in time. The task is challenging because of the complex characteristics of time-series, which are messy, stochastic, and often without proper labels. This prohibits training supervised models because of lack of labels and a single model hardly fits different time series. In this paper, we propose a solution to address these issues. We present an automated model selection framework to automatically find the most suitable detection model with proper parameters for the incoming data. The model selection layer is extensible as it can be updated without too much effort when a new detector is available to the service. Finally, we incorporate a customized tuning algorithm to flexibly filter anomalies to meet customers' criteria. Experiments on real-world datasets show the effectiveness of our solution.

cs.LG↗

arXiv2020

DeGNN: Characterizing and Improving Graph Neural Networks with Graph Decomposition

Despite the wide application of Graph Convolutional Network (GCN), one major limitation is that it does not benefit from the increasing depth and suffers from the oversmoothing problem. In this work, we first characterize this phenomenon from the information-theoretic perspective and show that under certain conditions, the mutual information between the output after $l$ layers and the input of GCN converges to 0 exponentially with respect to $l$. We also show that, on the other hand, graph decomposition can potentially weaken the condition of such convergence rate, which enabled our analysis for GraphCNN. While different graph structures can only benefit from the corresponding decomposition, in practice, we propose an automatic connectivity-aware graph decomposition algorithm, DeGNN, to improve the performance of general graph neural networks. Extensive experiments on widely adopted benchmark datasets demonstrate that DeGNN can not only significantly boost the performance of corresponding GNNs, but also achieves the state-of-the-art performances.

cs.LG↗

arXiv2020

Customized Graph Embedding: Tailoring Embedding Vectors to different Applications

Graph is a natural representation of data for a variety of real-word applications, such as knowledge graph mining, social network analysis and biological network comparison. For these applications, graph embedding is crucial as it provides vector representations of the graph. One limitation of existing graph embedding methods is that their embedding optimization procedures are disconnected from the target application. In this paper, we propose a novel approach, namely Customized Graph Embedding (CGE) to tackle this problem. The CGE algorithm learns customized vector representations of graph nodes by differentiating the importance of distinct graph paths automatically for a specific application. Extensive experiments were carried out on a diverse set of node classification datasets, which demonstrate strong performances of CGE and provide deep insights into the model.

cs.LG↗

arXiv2020

Deeper Insights into Weight Sharing in Neural Architecture Search

With the success of deep neural networks, Neural Architecture Search (NAS) as a way of automatic model design has attracted wide attention. As training every child model from scratch is very time-consuming, recent works leverage weight-sharing to speed up the model evaluation procedure. These approaches greatly reduce computation by maintaining a single copy of weights on the super-net and share the weights among every child model. However, weight-sharing has no theoretical guarantee and its impact has not been well studied before. In this paper, we conduct comprehensive experiments to reveal the impact of weight-sharing: (1) The best-performing models from different runs or even from consecutive epochs within the same run have significant variance; (2) Even with high variance, we can extract valuable information from training the super-net with shared weights; (3) The interference between child models is a main factor that induces high variance; (4) Properly reducing the degree of weight sharing could effectively reduce variance and improve performance.

cs.LG↗

arXiv2019

TextNAS: A Neural Architecture Search Space tailored for Text Representation

Learning text representation is crucial for text classification and other language related tasks. There are a diverse set of text representation networks in the literature, and how to find the optimal one is a non-trivial problem. Recently, the emerging Neural Architecture Search (NAS) techniques have demonstrated good potential to solve the problem. Nevertheless, most of the existing works of NAS focus on the search algorithms and pay little attention to the search space. In this paper, we argue that the search space is also an important human prior to the success of NAS in different applications. Thus, we propose a novel search space tailored for text representation. Through automatic search, the discovered network architecture outperforms state-of-the-art models on various public datasets on text classification and natural language inference tasks. Furthermore, some of the design principles found in the automatic network agree well with human intuition.

cs.LG↗

arXiv2019

A Code Injection Method for Rapid Docker Image Building

Docker images are composed of multiple layers, each of which contains a set of instructions, and an archive of files. Layers allow Docker to separate a large build task into smaller ones, such that when a part of the program is changed, only the corresponding layer needs to be changed. Yet the current implementation has major inefficiencies that make the rebuilding of an image unnecessarily slow when changes in bottom layers are required: uneven content distribution amongst layers, the need to rebuild an entire layer during update, and the rebuild fall-throughs in many cases. In this paper, we propose a code injection method that overcomes these inefficiencies by targeting only the changed layer and then bypassing the layer's content checksum. This process is developed specifically for an interpreted language such as Python, where changes can be detected explicitly via text diff tools and run as-is without compilation. We then demonstrate that this method can accelerate the rebuild time, effectively reducing the O(n) where n = size of layer rebuild time to O(1). Whereas for compiled languages, literal code injection cannot guarantee integrity in compiled machine code. Expanding on the same code injection principle, multi-layer targeted code injection will be addressed in a future discussion.

cs.DC↗

arXiv2019

Adapting a Container Infrastructure for Autonomous Vehicle Development

In the field of Autonomous Vehicle (AV) development, having a robust yet flexible infrastructure enables code to be continuously integrated and deployed, which in turn accelerates the rapid prototyping process. The platform-agnostic and scalable container infrastructure, often exploited by developers in the cloud domain, presents a viable solution addressing this need in AV development. Developers use tools such as Docker to build containers and Kubernetes to setup container networks. This paper presents a container infrastructure strategy for AV development, discusses the scenarios in which this strategy is useful and performs an analysis on container boundary overhead, and its impact on a Mix Critical System (MCS). An experiment was conducted to compare both operation runtime and communication delay of running a Gaussian Seidel Algorithm with I/O in four different environments: native OS, new container, existing container, and nested container. The comparison reveals that running in containers indeed adds a delay to signal response time, but behaves more deterministically and that nested container does not stack up delays but makes the process less deterministic. With these concerns in mind, the developers may be more informed when setting up the container infrastructure, and take full advantage of the new infrastructure while avoiding some common pitfalls.

cs.SE↗

arXiv2019

Developing a Process in Architecting Microservice Infrastructure with Docker, Kubernetes, and Istio

As an application usage grows, its owner scales up vertically by replacing old machines with more powerful ones. This methodology is expensive and leads to resource waste. In response to the business needs, internet giants have developed the microservice architecture, which lets developers divide up their application into smaller units that can be hosted on multiple machines, thus enabling horizontal scale up. We propose a triphasic incremental process to transform a traditional application into a microservice application that guarantees stability during the operation. Then we demonstrated such methodology in a prototype microservice application based on an existing monolithic application. First, the developer splits a monolithic application into atomic services and aggregated services. Second, these services are packaged, containerized, and then deployed on Kubernetes. During this stage, Istio is deployed on the Kubernetes cluster to establish pod level communications, delegate traffic flows and filter requests, and enable the autoscaler. Other external add-ons, such as database connections, are defined in service entry. In the last stage, we developed an algorithm guideline to minimize inter-service calls by compiling all needed calls into a list and perform one finalized call. Although it increases memory usage, it avoided the wait time incurred during interservice calls. We then investigated managing configurations using config maps, recommended a pipeline being developed to perform automatic rollover.

cs.DC↗

arXiv2019

Time-Series Anomaly Detection Service at Microsoft

Large companies need to monitor various metrics (for example, Page Views and Revenue) of their applications and services in real time. At Microsoft, we develop a time-series anomaly detection service which helps customers to monitor the time-series continuously and alert for potential incidents on time. In this paper, we introduce the pipeline and algorithm of our anomaly detection service, which is designed to be accurate, efficient and general. The pipeline consists of three major modules, including data ingestion, experimentation platform and online compute. To tackle the problem of time-series anomaly detection, we propose a novel algorithm based on Spectral Residual (SR) and Convolutional Neural Network (CNN). Our work is the first attempt to borrow the SR model from visual saliency detection domain to time-series anomaly detection. Moreover, we innovatively combine SR and CNN together to improve the performance of SR model. Our approach achieves superior experimental results compared with state-of-the-art baselines on both public datasets and Microsoft production data.

cs.LG↗

arXiv2019

A Right-of-Way Based Strategy to Implement Safe and Efficient Driving at Non-Signalized Intersections for Automated Vehicles

Non-signalized intersection is a typical and common scenario for connected and automated vehicles (CAVs). How to balance safety and efficiency remains difficult for researchers. To improve the original Responsibility Sensitive Safety (RSS) driving strategy on the non-signalized intersection, we propose a new strategy in this paper, based on right-of-way assignment (RWA). The performances of RSS strategy, cooperative driving strategy, and RWA based strategy are tested and compared. Testing results indicate that our strategy yields better traffic efficiency than RSS strategy, but not satisfying as the cooperative driving strategy due to the limited range of communication and the lack of long-term planning. However, our new strategy requires much fewer communication costs among vehicles.

cs.RO↗